We present a system to perform video analysis in the context of traffic surveillance's application. A training step is performed to estimate the scene's geometry and global information about the motion that occurs in the scene. Lanes boundaries, depth and motion information given by the initialization step are used to assist the vehicles' segmentation and to correct eventual errors.
Published in:
Image and Video Technology (PSIVT), 2010 Fourth Pacific-Rim Symposium on
Date of Conference: 14-17 Nov. 2010